Calibrated Trust in Dealing with LLM Hallucinations - Qualitative Survey Dataset

Calibrated Trust in Dealing with LLM Hallucinations - Qualitative Survey Dataset Overview This dataset originates from a qualitative online survey conducted between December 6, 2024, and January 8, 2025. The study explored user experiences with ChatGPT as an example of a large language model (LLM), focusing on the use context, awareness of AI-generated hallucinations, and trust-related behavior. Survey Details Data collection period: December 6, 2024 – January 8, 2025 Total participants: 192 Language: German Format: Online survey Content The survey included both closed-ended (e.g., Yes/No) and open-ended questions, allowing respondents to elaborate on their experiences. Key topics addressed in the survey: Frequency and type of LLM usage Awareness of LLM hallucinations (factually incorrect outputs) Personal experiences with LLM hallucinations Effects of hallucinations on trust in LLMs Whether LLMs are used as a support tool or a primary information source Strategies for information verification of the participants Basic demographic background (e.g., age, education, employment) Files Included data_survey_anonym.csv / data_survey_anonym.xlsx: Anonymized dataset of all survey responses, including both structured data and qualitative comments. Personally identifiable information has been removed. questionnaire.pdf / questionnaire.mhtml: Full version of the original questionnaire (in German), outlining all survey items and instructions shown to participants. Methodological Notes Participants were recruited online and responded voluntarily and anonymously. Even where questions were binary (e.g., Yes/No), participants were encouraged to elaborate through optional text fields. This approach allowed for rich qualitative data alongside basic metrics. No coding or interpretation is included in this dataset. For further analysis and discussion of the results, please refer to the corresponding publication.